Human Health Risk AssessmentQMRA within holistic AMR management

Nicholas ASHBOLTProfessor and Alberta Innovates Translational Research Chair in Waterborne Disease Prevention – Ashbolt@Ualberta.ca

PACCARB, PANEL SESSION 3: Public Health Risk, Impact, and Global ImplicationsSheraton Columbus Hotel at Capitol Square, September 26th, 2018

• University of Alberta - Founded in 190838,000 students from 140 countries, with 400 undergraduate and 500 graduate programs across 18 faculties

Ranked 90th in the 2018 QA World University RankingsOne of the top 5 research intensive universities in Canada

• School of Public Health, split from Medicine in 2006, first US accredited SPH in Canada• Three existing academic units provide the core components:

• Department of Public Health Sciences• Alberta Centre for Injury Control• Centre for Health Promotion Studies

Students• Some 300 graduate students (14 % distance & 15 % international)• 3 degree programs: MPH, MSc, PhD

Staff• 37 Core faculty members , 244 Research and support staff • 57 Jointly appointed or cross-appointed faculty members and 13 Postdoctoral fellowsAnnually• leverage over $16 million in research funding• lead or participates in more than 150 research projects• 175 peer-reviewed articles in academic journals• engage in > 200 knowledge exchange activities

Quantifying drivers of Environ AMR• Mitigating the risks of antibiotic resistance requires linking

various data reported on drivers of antibiotic resistance in humans arising from human, animal & environmental reservoirs

• Needs a One Health approach, plus re-framing to manage risk• Cochrane systematic review: selected 565 studies out of 2819 titles1

• Odds ratios of AMR 2-4x from antibiotic exposure, underlying disease & invasive procedures, from 88 risk factors studied; yet 1

• Food/water-related transmissions frequent 1

• Feedlot beef study suggests environ more than meat2

• Hence, environ important but how to link information together?1Chatterjee et al. (2018) Lancet Infect Dis doi.org/10.1016/S1473-3099(18)30296-22Noyes et al. (2016) eLife 5, e1319; Weinroth et al. (2018) Appl Env Microbiol 84 e00610-18 4

Environmental resistome is vast, ancient, mobilizable1

• Ample evidence that resistance elements currently circulating in pathogens belong to gene families with origins in the environment

• Hypothesis: use of antibiotics in bulk provides a strong selection for stochastic capture of environmental resistance genes, e.g.

• mobile genetic elements (MGE) eventually acquired by pathogens• Frequency of these capture events correlates with gene diversity & environ burden

• Goal: manage hot-spots where mixing of pathogens, mobile elements, and resistance genes is higher (manure, wastewater treatment, sediments etc.) – but what level?1Surette & Wright (2017) Annu Rev Microbiol 71, 309-330 5

QMRA process (U.S. EPA 2014; WHO 2016)

EPA (2014). Microbiological Risk Assessment (MRA) Tools, Methods, and Approaches for Water MediaWHO (2016). Quantitative Microbial Risk Assessment: Application for Water Safety Management World Health Organization: Geneva www.who.int/water_sanitation_health/publications/qmra/en/

Using scientific evidence to prioritize risk management

6

Petterson & Ashbolt (2016) J Wat Health 4(4), 571-589WHO (2016) Quantitative Microbial Risk Assessment, Geneva

QMRA - defined treatment requirements

7

Virus, bacteria & parasite levels for safe uses.

Drives regulations then surrogates to demonstrate

pathogen reductions at control points -

However AMR, as for saprozoic pathogens,

need additional controls

7

Environmental pathways of AMR

8Ashbolt et al. (2018) Antimicrobial Resistance…Global Water Pathogens Project; waterpathogens.org 8

Key questions when undertaking QMRA for AMR1

1. Which reference pathogens, ARGs or resistome: (WHO 2017) priority list ?2. Environmental pathways of exposure or just deal with hot-spots?

• AMR is a One Health problem, but has local, national & international aspects

3. Dose-response?• Does AMR change the probability of infection or just illness/treatment time and

therefore health burden (need DALY/QALY estimates)?• Some quantitative evidence for increased health burden

• How much HGT occurs on/within human host from environmental bacteria?

4. Benchmark risk – annual 10-4 infection, 10-6 DALY or Utility function?• Would log-removal targets work at treatment given episodic & environ pathways?• How to address (what level is OK) given post treatment ‘hot-spots’?

1Ashbolt et al. (2013) Environ Health Perspect 121(9), 993-1001 9

How to capture missing data & management links?• Traditionally from groups of experts & published literature• Which has evolved into broader stakeholder groups to provide

local, scientific and a more comprehensive understanding of complex and dynamic socio-ecological systems and processes1

• Nonetheless, many limitations with stakeholder participation1

• To reduce limitations, the process must be institutionalised, creating organisational cultures spanning one-health to facilitate processes where goals are negotiated & outcomes are necessarily uncertain

• In part, models aid in aggregating the collection of missing data• Specific use of Conceptual Models & Bayesian Networks (Bayes Nets)

1Reed (2008) Biological Conservation, 141, 2417-2431 10

Bayes Nets vs Multi-Criteria Decision Aiding (MCDA)1

• “Weighing available evidence in the process of decision-making is unavoidable, yet it is one step that routinely raises suspicions:

• what evidence, its weigh, and whose thumb may be tipping the scales?

• U.S. National Res Council critiqued ‘weight of evidence’ (WoE) as too vague and detractive for chemical risk assessment, so for AMR?

• How to move from qualitative, vague & controversial methods?• Bayesian WoE or MCDA?

• Inexact science of converting existing health/infection knowledge into risk management decisions and policies

• Noting ‘colloquial WoE use’ is the problem 1

1Linkov et al. (2015) ALTEX 32 (1), 3-8 11

Bayes net (BN)1

1Judea Pearl (2009) Causal inference in statistics: An overview. Statistics Surveys 3, 96-146Pearl & Russell (2001) Bayesian Networks, http://ftp.cs.ucla.edu/pub/stat_ser/R277.pdf

• Form of directed acyclic graph (DAG), where variables are X4 represented by nodes & linking arrows for causality

• The DAG provides an initial visual representation of relationships of influence among the variables (from Conceptual diagram)

• Each node or variable has a number of user-defined states that can be qualitative, discrete or continuous

• e.g., ‘high/low’, ‘true/false’, ‘≤ 104 / > 104 gene copies・gram-1’• Each of the variable states is assigned a conditional probability, derived

from published empirical data, models, simulations, or expert opinion• Use of Noisy-based logic functions (multivaled deterministic logic schemes)

12

X1

X2

X5

X3 Rain eventEffluentdischarge

MobilizedARB

AMR-infection

Season

Burden of disease: Use of Disability-adjusted life year (DALY) or the Utility function in Bayesian modelling?

• DALY is a common health metric• Adopted by WHO and many other health agencies, for:1) the 'positive' exercise of measuring the burden of disease; and 2) The 'normative' exercise of resource allocation

• However, the implications of age-weighting, blindness towards equity and discounting are unacceptable to some1

• Does not distinguish between measuring burden of disease and allocation of resources – as information sets are quite different between the two

• Allocating resources by aggregate DALY-minimisation is inequitable• Therefore suggest use of ‘Utility’ function (Von Neumann Morgenthau Utility theorem)2

• Utilities can be used as the quality-adjustment weights in DALYs or QALYs• Variation on cost-effectiveness analysis known as cost-utility analysis

1Anad & Hanson (1997) J Health Econ 16 (6), 685-702Whitehead & Ali (2010) Br Med Bull 96, 5-21

2Torrance & Ferny (1989) Int J Technol Assess Health Care 5 (4), 559-575 13

Unknowns: environ exposure, growth & impacts of AMR bacteria1 – what Bayes Nets could address

Water & wastewater

Fertilizer & food safety

Animal production& companion

animals

Phama & hospital effluents

Management programs & policies

Human Exposures

Aquatic/soil environment

Recreation/fishing

Crops/ fruit

Irrigation aerosols

FoodCompanion

animals

Drinking water

Health care settings

U-bend wastewater

outlets

Environmental exposures (setting priorities & limits)

Travel

recreation

Banknotes & other fomites

aerosols

ingestionHand-to-mouth

1Ashbolt et al. (2013) Environ Health Perspect 121(9), 993-1001 14

Conclusion: QMRA-driven AMR risk management should create and protect value/utility

• AMR management is all about risk minimization and health value/utility maximisation

• As a result we are faced with the same task as economists• But microbiologists generally don’t talk or think about economics

• Arguable we have a universal utility metric, the µDALY• Next logical step in QMRA – along with QMRA fitting risk tools

(ISO 31010/ANSI Z690) is introduce Utility into AMR management• Bayes Nets and Bayes theory can facilitate the whole process by

providing a single framework – focused on informing management15